A sports training scheme generation method and system for limb feature recognition

Abstract

The application relates to a sports training scheme generation method and system for limb feature recognition, and relates to the field of sports training. A skeleton topological portrait and a body shape proportion portrait of a target user are collected. A joint number list and a joint standard time sequence coordinate list of a to-be-trained movement type meeting the portrait are searched. When the target user is trained, a joint time sequence deviation coordinate list is obtained by comparing a joint training time sequence coordinate list with the joint standard time sequence coordinate list. A time proportion list of a time point at which a coordinate deviation distance is greater than or equal to a deviation distance threshold value is counted by traversing the joint time sequence deviation coordinate list. A joint number set at a time point at which a time proportion is greater than or equal to a time proportion threshold value is extracted, an abnormal action set is obtained by inputting the action matching library to execute abnormal action configuration, and a sports training scheme is configured. The technical problems that sports training ignores individual body differences, resulting in insufficient monitoring accuracy and poor individualization of sports training schemes are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of sports training, and in particular to a sports training scheme generation method and system based on limb feature recognition. BACKGROUND

[0002] In the existing field of sports training, the formulation of training schemes is often based on general standards and experience, ignoring the physical differences between individuals. Such universal training schemes have many problems in practical application, the most prominent of which are insufficient monitoring accuracy and poor individualization. First, due to differences in individual skeletal structure, body proportions, and movement habits, different people will have different joint movement trajectories, timing, and force when performing the same movement. The existing training schemes do not take these differences into account, resulting in an inability to accurately monitor individual training during actual training, which not only affects the evaluation of training effectiveness, but also may pose a potential threat to the safety of individuals. Second, due to the existence of individual differences, universal training schemes often cannot meet the individualized training needs of individuals. For some individuals, the training scheme may be too simple and unable to effectively improve their athletic ability; for others, the training scheme may be too complex or too intense, easily leading to sports injuries or creating a negative attitude. This lack of individualization not only affects the improvement of training effectiveness, but also may reduce the enthusiasm and participation of individuals in training. SUMMARY

[0003] The present application provides a sports training scheme generation method and system based on limb feature recognition to solve the technical problems of ignoring individual physical differences, resulting in insufficient monitoring accuracy and poor individualization of sports training schemes in the prior art.

[0004] The technical solution of the present application to solve the above technical problems is as follows:

[0005] In a first aspect, the present application provides a sports training scheme generation method based on limb feature recognition, comprising: collecting a skeletal topological portrait and a body proportion portrait of a target user; retrieving a joint number list and a joint standard timing coordinate list of a to-be-trained movement type that meets the skeletal topological portrait and the body proportion portrait; when the target user is training, extracting a joint training timing coordinate list, comparing it with the joint standard timing coordinate list, and obtaining a joint timing deviation coordinate list; traversing the joint timing deviation coordinate list, and counting a time proportion list of time points with a coordinate deviation distance greater than or equal to a deviation distance threshold; based on the time proportion list, extracting a joint number set with a time proportion greater than or equal to a time proportion threshold, inputting an action matching library to perform abnormal action configuration, obtaining an abnormal action set, and configuring a sports training scheme.

[0006] In a second aspect, the present application provides a sports training scheme generation system for body feature recognition, comprising: an image acquisition module, configured to acquire a skeletal topology image and a body shape proportion image of a target user; a data retrieval module, configured to retrieve a joint number list and a joint standard time sequence coordinate list of a to-be-trained movement type that meets the skeletal topology image and the body shape proportion image; a deviation comparison module, configured to extract a joint training time sequence coordinate list when the target user is training, compare the joint training time sequence coordinate list with the joint standard time sequence coordinate list, and obtain a joint time sequence deviation coordinate list; a statistical analysis module, configured to traverse the joint time sequence deviation coordinate list, and statistically analyze a time point proportion list in which a coordinate deviation distance is greater than or equal to a deviation distance threshold; a scheme configuration module, configured to extract a joint number set in which a time point proportion is greater than or equal to a time point proportion threshold based on the time point proportion list, input an action matching library to perform abnormal action configuration, obtain an abnormal action set, and configure a sports training scheme.

[0007] The present application has the beneficial effects that: by acquiring a skeletal topology image and a body shape proportion image of a target user, and retrieving a corresponding joint standard time sequence coordinate list, real-time monitoring of the training of the target user is realized, abnormal actions are identified by comparing joint training time sequence coordinates with standard time sequence coordinates, and a personalized sports training scheme is configured accordingly, so that the individualization degree and monitoring accuracy of the training scheme are improved. BRIEF DESCRIPTION OF DRAWINGS

[0008] Figure 1 A flowchart of a sports training scheme generation method for body feature recognition provided by the present application is shown.

[0009] Figure 2 A structural diagram of a sports training scheme generation system for body feature recognition provided by the present application is shown.

[0010] Marked with: image acquisition module 11, data retrieval module 12, deviation comparison module 13, statistical analysis module 14, scheme configuration module 15. DETAILED DESCRIPTION

[0011] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0012] In the description of the present application, the terms "first", "second" are only for descriptive purpose, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "a plurality of" is two or more, unless otherwise specifically limited.

[0013] In the description of the present application, the term "for example" is used to indicate "as an example, illustration or description". Any embodiment described as "for example" in the present application is not necessarily interpreted as more preferred or more advantageous than other embodiments. The following description is given to enable any person skilled in the art to implement and use the present application. In the following description, details are listed for the purpose of explanation. It should be understood that those skilled in the art can recognize that the present application can be implemented without using these specific details. In other examples, well-known structures and processes will not be described in detail to avoid unnecessary details making the description of the present application obscure. Therefore, the present application is not intended to be limited to the embodiments shown, but is consistent with the broadest scope consistent with the principles and characteristics disclosed.

[0014] Embodiment one:

[0015] As shown in the figure, the embodiment of the present application provides a sports training scheme generation method for limb feature recognition, comprising: Figure 1

[0016] S10: Collecting the skeletal topological portrait and the body shape ratio portrait of the target user.

[0017] S20: Retrieving the joint number list and the joint standard time sequence coordinate list of the training type that meets the skeletal topological portrait and the body shape ratio portrait.

[0018] S30: When the target user trains, extracting the joint training time sequence coordinate list, comparing with the joint standard time sequence coordinate list, and obtaining the joint time sequence deviation coordinate list.

[0019] S40: Traversing the joint time sequence deviation coordinate list, and counting the time proportion list of the coordinate deviation distance greater than or equal to the deviation distance threshold.

[0020] S50: Based on the time proportion list, extracting the joint number set with time proportion greater than or equal to the time proportion threshold, inputting the action matching library to execute the abnormal action configuration, obtaining the abnormal action set, and configuring the sports training scheme.

[0021] ​For example, the bone topology image and the body proportion image of the target user are collected. The bone detection image and the body detection image of the target user in a preset posture are obtained through image processing technology, which can clearly show the bone structure and body shape characteristics of the target user. The bone topology image refers to the unique bone structure feature map of the target user, which is constructed by identifying the distribution position of the joint through the active joint calibration library based on the bone detection image of the target user. It reflects the relative position relationship and overall shape of the bones. The body proportion image refers to the body shape characteristics of the target user, including height, limb length ratio, etc., which are extracted by identifying the limb proportion of the bone detection image of the target user. It embodies the body shape and proportion characteristics of the target user. In short, the bone topology image focuses on the structure and layout of the bones, while the body proportion image focuses on the shape and proportion of the body. The two together constitute a comprehensive description of the body characteristics of the target user, providing an important basis for subsequent training monitoring and personalized training program development.

[0022] Specifically, taking a basketball player as an example, its bone topology image may show relatively slender limbs and wide shoulders, which is consistent with its sports needs, because such body shape is more conducive to jumping, shooting and defending. The body proportion image will further refine this feature, such as the specific length ratio of the limbs, the ratio of the torso to the lower limbs, etc., which will provide an important basis for the subsequent retrieval of joint standard time sequence coordinate list and identification of abnormal motion. This collection process not only ensures the accuracy and completeness of the data, but also lays a solid foundation for the subsequent development of personalized training programs.

[0023] Further, the joint number list and the joint standard time sequence coordinate list of the to-be-trained sports type that meet the bone topology image and the body proportion image are retrieved. The to-be-trained sports type refers to the sports project or motion type that the target user plans to train, such as running, swimming, basketball shooting, etc. The joint number list refers to the numbering of all key joints involved in the sports according to the to-be-trained sports type, which is used to identify and distinguish different joints for subsequent data processing and analysis. For example, in the basketball shooting motion, the joint number list may include the numbering of shoulder joint, elbow joint, wrist joint, etc. The joint standard time sequence coordinate list refers to the set of standard time sequence coordinates that each key joint should reach when performing the to-be-trained sports type. These coordinates represent the position and motion trajectory of the joint in space, as well as the motion state at different time points. By comparing these standard time sequence coordinates with the joint time sequence coordinates of the target user during actual training, the training effect of the target user can be evaluated, abnormal motions can be identified, and personalized sports training programs can be configured accordingly.

[0024] Specifically, the target user's skeletal structure and body shape characteristics are used to search for similar user samples in the first-level qualified user sample group. The first-level qualified user sample group refers to a collection of user samples whose skeletal structure and body shape characteristics are similar to those of the target user, and whose motion data has high reference value and accuracy. These samples have reached a high scoring standard (i.e., first-level scoring) in both skeletal topology portraits and body shape proportion portraits. Therefore, they can be used to retrieve and generate joint numbering lists and joint standard time sequence coordinate lists for the target user that conform to the individual characteristics of the target user for the to-be-trained motion type. If the number of first-level qualified user sample groups is insufficient, the second-level qualified user sample group is further traversed to expand the search range. In this process, high-frequency fitting is performed on the joint time sequence coordinates in the qualified user sample group using all joint numbering lists, thereby obtaining the joint standard time sequence coordinate list that the target user should achieve when performing a specific to-be-trained motion type.

[0025] For example, taking an athlete who wants to improve his basketball skills as an example, based on the characteristics such as long limbs and broad shoulders shown in the skeletal topology portrait, as well as the height and limb length proportion information in the body shape proportion portrait, samples of basketball players with similar characteristics are searched for in the first-level or second-level qualified user sample group. Then, the joint time sequence coordinates of these samples when performing basketball training are extracted, and the joint standard time sequence coordinate list that the athlete should achieve when performing basketball training is obtained through high-frequency fitting. In this way, the coach can monitor and adjust the athlete's training according to this list to ensure that his movements meet the standards, thereby improving the training effect.

[0026] In detail, when the target user starts training, the joint training time sequence coordinate list is extracted in real time, and these coordinate data reflect the actual positions and movement trajectories of each joint of the target user during the movement. Subsequently, the real-time extracted joint training time sequence coordinates are compared with the joint standard time sequence coordinate list. The joint standard time sequence coordinate list is obtained through high-frequency fitting based on a large amount of motion data of similar users, and represents the ideal positions and movement trajectories that the joints should achieve when performing a specific motion. In the comparison process, the position deviation of each joint at each time point is calculated, thereby generating a joint time sequence deviation coordinate list. This list records in detail the deviation of each joint of the target user relative to the standard time sequence coordinates during the training process. For example, if the real-time position of the elbow joint of the target user deviates from the standard time sequence coordinates when performing basketball shooting training, this deviation will be reflected in the joint time sequence deviation coordinate list. Through this process, the coach and trainer can clearly understand the movement accuracy of the target user during the training process, and timely discover and correct abnormal movements, thereby improving the training effect.

[0027] Next, after obtaining the list of joint timing deviation coordinates, the list is further traversed to analyze the deviation distance of each joint at each time point. A deviation distance threshold is set, which is determined based on sports science theory and actual training experience, to determine whether the joint deviation is within an acceptable range. When the deviation distance of a joint at a certain time point is greater than or equal to the deviation distance threshold, it is recorded.

[0028] By counting the number of times the deviation distance is greater than or equal to the deviation distance threshold, a time proportion list can be generated, which reflects the proportion of time that the joint deviation exceeds the acceptable range during the entire training process of the target user. For example, if the target user's elbow joint deviation distance exceeds the deviation distance threshold at multiple time points during basketball shooting training, these time points will be counted in the time proportion list, and the elbow joint's time proportion will be relatively high. Such analysis helps coaches and trainers better understand the target user's training situation, especially the accuracy and stability of joint movements, so that they can develop more targeted training plans and improve training effectiveness.

[0029] Finally, based on the generated time proportion list, joints with a time proportion greater than or equal to a preset time proportion threshold are further filtered out. These joint numbers represent joints with large deviations and inaccurate movements during the target user's training process. For example, if the target user's knee and ankle joints have a time proportion greater than the set threshold during a certain sports training, the numbers of these two joints will be extracted. Next, the extracted joint number set is input into the action matching library. The action matching library is a database containing a large number of standard and abnormal action templates. The library searches for abnormal action templates that match the joint number set. Through this process, the specific abnormal actions of the target user during training can be identified. After obtaining the abnormal action set, a personalized sports training plan can be configured based on the characteristics of these abnormal actions and the training needs of the target user. For example, if the target user is identified to have elbow and wrist joint abnormalities during basketball shooting training, the training plan may include specialized correction training for these two joints and exercises to strengthen related muscle strength to help the target user correct abnormal movements and improve training effectiveness.

[0030] In a preferred embodiment, the bone topology image and the body shape proportion image of the target user are collected, including: obtaining a bone detection map and a body shape detection map of a preset posture of the target user; receiving the to-be-trained exercise type from the user end, processing through an active joint calibration library to obtain a joint number list, wherein the active joint calibration library is pre-constructed for the user and stores a one-to-one correspondence between the exercise type and the joint number list; based on the joint number list, performing distributed position recognition on the bone detection map to obtain joint topology features, which are set as the bone topology image; based on the joint number list, performing limb proportion recognition on the bone detection map to obtain limb proportion features, which are set as the body shape proportion image.

[0031] Optionally, in the process of collecting the bone topology image and the body shape proportion image of the target user, the target user first needs to assume a preset posture, and the bone detection map and the body shape detection map are obtained through professional detection equipment. These detection maps can clearly show the skeletal structure and body shape features of the target user. Then, the information of the to-be-trained exercise type is received from the user end, and the pre-constructed active joint calibration library is used for processing. The active joint calibration library stores a one-to-one correspondence between the exercise type and the joint number list, which can quickly determine the joints that the target user needs to pay attention to when performing a specific exercise. For example, if the target user wants to perform basketball training, the active joint calibration library will provide a joint number list related to basketball exercises, such as shoulder joints, elbow joints, wrist joints, etc. Based on this joint number list, the distributed position recognition can be performed on the bone detection map to extract the relative position relationship of each joint in the skeletal structure, thereby constructing the bone topology image of the target user, which can intuitively reflect the skeletal morphology and joint distribution features of the target user. At the same time, the limb proportion recognition is performed on the bone detection map based on the joint number list. By analyzing the length, width, and other proportion relationships of each limb in the bone detection map, the limb proportion features of the target user, i.e., the body shape proportion image, can be obtained. This image can show the body morphology and proportion features of the target user, providing an important reference for subsequent training plan formulation. Taking a user who wants to perform yoga training as an example, by collecting the bone topology image and the body shape proportion image, the flexibility of the skeletal structure and the coordination of the limb proportions of the user can be understood, thereby formulating a more personalized yoga training plan for the user to improve the training effect.

[0032] In a preferred embodiment, the retrieving of the joint number list and the joint standard time sequence coordinate list of the to-be-trained movement type satisfying the skeleton topology image and the body shape proportion image comprises: retrieving a first-level score qualified user sample group satisfying the skeleton topology image and the body shape proportion image, wherein any one of the first-level score qualified user sample group has a first-level skeleton topology image and a first-level body shape proportion image; when the number of the first-level score qualified user sample group is less than or equal to a statistical number threshold, traversing the first-level score qualified user sample group to retrieve a second-level score qualified user sample group satisfying the first-level skeleton topology image and the first-level body shape proportion image; and traversing the joint number list to perform high-frequency fitting on the joint time sequence coordinates of the first-level score qualified user sample group and the second-level score qualified user sample group to obtain the joint standard time sequence coordinate list.

[0033] In detail, in the retrieving of the joint number list and the joint standard time sequence coordinate list of the to-be-trained movement type satisfying the target user skeleton topology image and the body shape proportion image, firstly, focus on the first-level score qualified user sample group, the skeleton topology image and the body shape proportion image of these user samples all reach the first-level standard, which means that their body structure and proportion are highly similar to the target user, and therefore their movement data has high reference value. For example, if the target user is a dancer with a slender body and flexible joints, look for dancer samples with similar skeleton topology and body shape proportion characteristics in the first-level score qualified user sample group.

[0034] However, if the number of the first-level score qualified user sample group is insufficient to meet the retrieval requirements, i.e., less than or equal to a preset statistical number threshold, further expand the retrieval range, traverse the first-level score qualified user sample group, and continue to retrieve a second-level score qualified user sample group satisfying the first-level skeleton topology image and the first-level body shape proportion image. These second-level samples, although with slightly lower scores, still have high similarity and can provide more useful data for retrieval.

[0035] After obtaining a sufficient number of qualified user samples, traverse the joint number list to perform high-frequency fitting on the joint time sequence coordinates of these samples. This process aims to find the standard time sequence coordinates that each joint should reach under a specific to-be-trained movement type. For example, in dance training, the joint time sequence coordinates of a large number of dancers can be analyzed to obtain, through high-frequency fitting, how each joint should coordinate movement to achieve the best effect in a specific dance movement. Finally, generate a joint standard time sequence coordinate list according to the results of high-frequency fitting, which not only reflects the ideal movement trajectory of each joint under the to-be-trained movement type, but also serves as a reference standard for the target user during training, helping to correct movement deviations and improve training effectiveness.

[0036] In a preferred embodiment, the joint number list is traversed, and high-frequency fitting of joint time-series coordinates is performed on the primary score qualified user sample group and the secondary score qualified user sample group to obtain the joint standard time-series coordinate list, including: extracting a first joint number from the joint number list; extracting a first time-series coordinate set of the first joint number of the secondary score qualified user sample group; extracting a first time point coordinate set to an Nth time point coordinate set of the time-series coordinate set; traversing the first time point coordinate set to the Nth time point coordinate set to perform centralized trend analysis to obtain a first time point coordinate set value to an Nth time point coordinate set value; extracting a second time-series coordinate set of the first joint number of the primary score qualified user sample group; traversing the first time point coordinate set value to the Nth time point coordinate set value, and performing time point centralized trend analysis in combination with the second time-series coordinate set to obtain a first joint number standard time-series coordinate, which is added to the joint standard time-series coordinate list.

[0037] Specifically, the first joint number, such as the elbow joint number, is extracted from the joint number list. Then, for the secondary score qualified user sample group, all time sequence coordinates corresponding to the elbow joint number are extracted to form a first time sequence coordinate set, which contains the elbow joint coordinate data of all users in the secondary sample group at each time. Then, further extract the coordinate set at the first time from the first time sequence coordinate set, and the coordinate set at the Nth time, that is, group the coordinate data in chronological order. Next, the central tendency analysis is performed on each group of time coordinate sets, such as calculating the mean or median, to obtain the central value of the first time coordinate set to the Nth time coordinate set, which represents the average or typical position of the secondary sample group at each time on the joint number. Then, turn to the primary score qualified user sample group and extract the second time sequence coordinate set of the same joint number (such as the elbow joint). Since the user data quality of the primary sample group is higher, this step is to further refine and calibrate the standard time sequence coordinates based on the secondary sample group. Next, iterate through the first time coordinate central value to the Nth time coordinate central value calculated before, and perform central tendency analysis at the same time combined with the second time sequence coordinate set of the primary sample group. The purpose of this step is to fuse the data of the primary and secondary sample groups to obtain more accurate and reliable standard time sequence coordinates. For example, if the elbow joint coordinates of the primary sample group at a certain time are generally more concentrated and more consistent with the kinematics principle than those of the secondary sample group, the final standard time sequence coordinates will be more inclined to the data of the primary sample group. Finally, the standard time sequence coordinates of the joint number (such as the elbow joint) are obtained by analysis and added to the joint standard time sequence coordinate list. This process is repeated for each joint number in the joint number list until the complete joint standard time sequence coordinate list is generated. This list provides an important reference for subsequent personalized training scheme development, ensuring that the training is more scientific and effective.

[0038] In a preferred embodiment, based on the time proportion list, a joint number set with a time proportion greater than or equal to a time proportion threshold is extracted, an abnormal action configuration is performed on the action matching library to obtain an abnormal action set, and a sports training scheme is configured, including: step one: k-item number enumeration combination is performed on the joint number set to obtain a plurality of joint number combinations, wherein k is initially equal to 1, the total number of joint numbers ≥ k ≥ 1, and k is an integer; step two: based on the type of the to-be-trained movement, a first action joint number combination to an Mth action joint number combination is obtained from the action matching library; step three: iterate through the plurality of joint number combinations to match the first action joint number combination to the Mth action joint number combination to obtain a k-item abnormal action set, which is added to the abnormal action set; if k < the total number of joint numbers, k is incremented by one and the process returns to step one, otherwise, the abnormal action set is output.

[0039] An exemplary k-item number enumeration combination is performed on the extracted joint number set with a time proportion greater than or equal to the time proportion threshold. Enumeration combination is a mathematical method that refers to selecting a number of elements from a given set of elements according to certain rules and listing all possible combinations. Here, k is an integer, and its initial value is set to 1, and it satisfies the condition that the total number of joint numbers is greater than or equal to k, which is greater than or equal to 1. Through enumeration combination, a number of different joint number combinations can be generated, providing a rich data basis for subsequent action matching. For example, if the joint number set is {1, 2, 3} and the initial value of k is 1, then {1}, {2}, and {3} are generated first. Then, the value of k is increased, and combinations such as {1, 2}, {1, 3}, and {2, 3} are generated, until the value of k equals the total number of joint numbers. Enumeration combination can list all possible combinations, ensuring that no possibility is missed, and by combining according to certain rules, enumeration combination can generate combinations with logic and system.

[0040] Based on the type of exercise to be trained, the first action joint number combination to the Mth action joint number combination related to the exercise type are retrieved from the action matching library, which represent various normal and abnormal action patterns that may occur when performing a specific exercise. Then, the several joint number combinations generated previously are traversed and matched one by one with the action joint number combinations retrieved from the action matching library. In the matching process, joint number combinations corresponding to abnormal action patterns are identified and classified into k abnormal action sets, which reflect the types of abnormal actions that the target user may have during training.

[0041] To ensure comprehensiveness and accuracy, it is checked whether the value of k is less than the total number of joint numbers. If so, the value of k is incremented by one, and the system returns to step one to continue generating new joint number combinations and matching. This process is repeated until the value of k equals the total number of joint numbers.

[0042] After all possible joint number combinations have been traversed and matched, the complete abnormal action set is output, which contains all abnormal action types that the target user may have during training, providing an important reference for subsequent sports training program configuration. For example, if the target user's elbow joint and wrist joint time proportions exceed the threshold during basketball shooting training, "elbow joint overflexion" and "wrist joint varus" abnormal actions may be identified through this series of steps, and a targeted correction training program can be developed for the user accordingly.

[0043] In a preferred embodiment, the action matching library updating step comprises: collecting a plurality of preset action training logs of a preset movement type, wherein the plurality of preset action training logs have a plurality of joint number record combinations; according to the plurality of joint number record combinations, counting a plurality of joint number trigger frequency proportions; extracting joint numbers with trigger frequency proportions greater than or equal to a trigger frequency proportion threshold from the plurality of joint number trigger frequency proportions, adding the joint numbers into a preset action joint number combination, and binding the preset action joint number combination with the preset movement type, and storing the preset action joint number combination into the action matching library.

[0044] Preferably, the updating of the action matching library aims to ensure that the action templates in the library can accurately reflect the joint activity in actual movements. A plurality of preset action training logs of a preset movement type are collected, which record the joint activity data of athletes during a specific preset action. These logs contain a plurality of joint number record combinations, each of which represents the cooperative working state of each joint at a certain moment or during a certain action. According to these joint number record combinations, the trigger frequency proportions of a plurality of joint numbers can be counted. The trigger frequency proportion refers to the frequency of a joint number appearing in all record combinations, which reflects the importance and activity level of the joint in a specific preset action. For example, in the training logs of a basketball shooting action, the trigger frequency proportions of the elbow joint and the wrist joint may be relatively high, because these two joints play a key role in the shooting action.

[0045] Then, joint numbers with trigger frequency proportions greater than or equal to a trigger frequency proportion threshold are extracted, which represent the joints that are most active and critical during the preset action. These joint numbers are added to the preset action joint number combination and bound with the corresponding preset movement type. This combination of preset movement type and key joint numbers is stored in the action matching library. In this way, when subsequent users perform similar movements, the action matching library can quickly identify the user's action pattern and compare it with the preset standard action template, thereby providing more accurate training feedback and correction suggestions to the user. For example, if the action matching library has already stored the joint number combination of the basketball shooting action, when the user performs shooting training, the joint activity data of the user can be analyzed in real time and compared with the standard template in the library, and the user's action deviation can be discovered and corrected in a timely manner.

[0046] In a preferred embodiment, based on the time proportion list, a set of joint numbers with time proportions greater than or equal to a time proportion threshold is extracted, an abnormal action configuration is performed on an action matching library to obtain an abnormal action set, and a sports training scheme is configured, including: extracting a first abnormal action from the abnormal action set; extracting the trained number of times of the first abnormal action; combining the historical training log to count the average training number of times of the first abnormal action; when the trained number of times is less than the average training number of times, identifying the target user as a regular user; when the trained number of times is greater than or equal to the average training number of times, identifying the target user as an enhanced training user; and configuring a first abnormal action sports training scheme according to the regular user identification or the enhanced training identification, and adding it to the sports training scheme.

[0047] Specifically, after obtaining the abnormal action set, a first abnormal action is further extracted therefrom, and the trained number of times of the abnormal action is queried. The trained number of times reflects the correction and practice of the target user on this abnormal action in the past training. In order to more comprehensively evaluate the training needs of this abnormal action, the average training number of times of the abnormal action is counted in combination with the historical training log. This average number of times represents the training amount that all users usually need to face this abnormal action.

[0048] Next, according to the comparison result of the trained number of times and the average training number of times, the target user is identified. If the trained number of times is less than the average training number of times, it indicates that the target user has not sufficiently corrected and practiced this abnormal action, and therefore is identified as a regular user. This means that in the subsequent training scheme, the training intensity and content for this abnormal action will remain at a regular level. Conversely, if the trained number of times is greater than or equal to the average training number of times, it indicates that the target user has corrected and practiced this abnormal action more, but still has not completely mastered it or there are repeated occurrences, and therefore is identified as an enhanced training user. This means that in the subsequent training scheme, the training intensity and content for this abnormal action will be increased to strengthen the training effect. Finally, a corresponding sports training scheme is configured for the first abnormal action according to the regular user identification or the enhanced training identification, and is added to the overall sports training scheme. For example, if the target user is identified as a regular user, the training scheme for its first abnormal action (such as "knee joint inward buckling") may include some basic correction exercises and flexibility training. If the target user is identified as an enhanced training user, the training scheme may include higher intensity strength training and more complex action mode training to help it better master and correct this abnormal action.

[0049] The sports training scheme generation method for limb feature recognition provided by the embodiment of the present application has at least the following technical effects:

[0050] 1. By collecting the skeletal topology and body proportion profiles of target users, and combining them with the joint number list and standard time-series coordinate list of the type of exercise to be trained, highly personalized sports training plans can be generated. This precise customization not only considers the user's physical structure characteristics, but also combines the characteristics of the type of exercise, thereby ensuring that the training plan is more in line with the user's actual needs and improves the training effect.

[0051] 2. During training, by extracting the joint training time-series coordinate list in real time and comparing it with the standard joint time-series coordinate list, abnormal user movements can be dynamically identified, and a list of the percentage of times when the coordinate deviation distance is greater than or equal to the deviation distance threshold can be compiled. Based on this list, a set of key joint numbers can be further extracted and input into the action matching library to execute abnormal action configuration, thereby obtaining a set of abnormal actions and configuring corresponding training schemes. This dynamic identification and correction mechanism can promptly detect and correct user movement deviations, avoid the solidification of erroneous movements, and improve the accuracy and effectiveness of training.

[0052] 3. The system considers the user's training history and the number of times abnormal actions have been trained. By combining historical training logs, the average number of training sessions for abnormal actions is calculated. Based on the comparison between the number of trained sessions and the average number of training sessions, users are categorized as either regular users or those requiring intensive training. Based on this categorization, the system can intelligently adjust training intensity and content, providing more suitable training programs for users at different training stages. This intelligent adjustment mechanism ensures that the training program is neither too easy nor too demanding, thereby maintaining the user's training motivation and continuity.

[0053] Example 2:

[0054] like Figure 2 As shown, based on the same inventive concept as the method for generating sports training programs using limb feature recognition provided in Embodiment 1, this embodiment of the invention also provides a sports training program generation system using limb feature recognition, the system comprising:

[0055] The image acquisition module 11 is used to acquire the skeletal topology image and body proportion image of the target user.

[0056] The data retrieval module 12 is used to retrieve a list of joint numbers and a list of standard time-series coordinates of joints that satisfy the skeletal topology profile and the body proportion profile for the type of exercise to be trained.

[0057] The deviation comparison module 13 is used to extract the joint training time sequence coordinate list when the target user is training, and compare it with the joint standard time sequence coordinate list to obtain the joint time sequence deviation coordinate list.

[0058] The statistical analysis module 14 is configured to traverse the joint time sequence deviation coordinate list to obtain a time proportion list of time points at which the coordinate deviation distance is greater than or equal to the deviation distance threshold.

[0059] The scheme configuration module 15 is configured to extract a joint number set at time points at which the time proportion is greater than or equal to the time proportion threshold based on the time proportion list, perform abnormal action configuration on the action matching library to obtain an abnormal action set, and configure a sports training scheme.

[0060] Further, the image acquisition module 11 is further configured to perform the following steps:

[0061] The skeleton detection image and the body shape detection image of the preset posture of the target user are obtained, the type of the to-be-trained movement is received from the user end, the active joint calibration library is processed to obtain the joint number list, the active joint calibration library is pre-constructed for the user and stores a one-to-one correspondence between the type of movement and the joint number list, the distribution position of the skeleton detection image is recognized based on the joint number list to obtain joint topology features, and the skeleton topology image is set; the limb proportion of the skeleton detection image is recognized based on the joint number list to obtain limb proportion features, and the body shape proportion image is set.

[0062] Further, the data retrieval module 12 is further configured to perform the following steps:

[0063] The first-level score qualified user sample group that meets the skeleton topology image and the body shape proportion image is retrieved, wherein any one of the first-level score qualified user sample group has a first-level skeleton topology image and a first-level body shape proportion image; when the number of the first-level score qualified user sample group is less than or equal to a statistical number threshold, the first-level score qualified user sample group is traversed to retrieve a second-level score qualified user sample group that meets the first-level skeleton topology image and the first-level body shape proportion image; the first-level score qualified user sample group and the second-level score qualified user sample group are subjected to joint time sequence coordinate high-frequency fitting based on the joint number list to obtain the joint standard time sequence coordinate list.

[0064] Further, the data retrieval module 12 is further configured to perform the following steps:

[0065] extracting a first joint number from the joint number list; extracting a first time sequence coordinate set of the first joint number of the secondary score qualified user sample group; extracting a first time point coordinate set to an Nth time point coordinate set of the time sequence coordinate set; performing centralized trend analysis on the first time point coordinate set to the Nth time point coordinate set, obtaining a first time point central value to an Nth time point central value; extracting a second time sequence coordinate set of the first joint number of the primary score qualified user sample group; performing simultaneous time point centralized trend analysis on the first time point central value to the Nth time point central value in combination with the second time sequence coordinate set, obtaining a first joint number standard time sequence coordinate, and adding the first joint number standard time sequence coordinate into the joint standard time sequence coordinate list.

[0066] Further, the scheme configuration module 15 is further configured to perform the following steps:

[0067] Step one: performing k-item number enumeration combination on the joint number set to obtain a plurality of joint number combinations, wherein k is initially equal to 1, the total number of joint numbers is greater than or equal to k and k is greater than or equal to 1, and k is an integer; step two: based on the to-be-trained motion type, obtaining a first motion joint number combination to an Mth motion joint number combination from the action matching library; step three: traversing the plurality of joint number combinations, matching the first motion joint number combination to the Mth motion joint number combination to obtain a k-item abnormal action set, and adding the k-item abnormal action set into the abnormal action set; wherein if k is less than the total number of joint numbers, k is incremented by one, and the process returns to step one, otherwise, the abnormal action set is output.

[0068] Further, the scheme configuration module 15 is further configured to perform the following steps:

[0069] Collecting a plurality of preset action training logs of a preset motion type, wherein the plurality of preset action training logs have a plurality of joint number record combination; according to the plurality of joint number record combination, a plurality of joint number trigger frequency proportions are counted; extracting joint numbers with trigger frequency proportions greater than or equal to a trigger frequency proportion threshold from the plurality of joint number trigger frequency proportions, adding the joint numbers into a preset action joint number combination bound with the preset motion type, and storing the joint numbers into the action matching library.

[0070] Further, the scheme configuration module 15 is further configured to perform the following steps:

[0071] According to the abnormal action set, a first abnormal action is extracted; a trained number of the first abnormal action is extracted; an average training number of the first abnormal action is counted in combination with a historical training log; when the trained number is less than the average training number, a target user is identified as a regular user; when the trained number is greater than or equal to the average training number, the target user is identified as an enhanced training user; and a first abnormal action sports training scheme is configured according to the regular user identification or the enhanced training identification, and is added to the sports training scheme.

[0072] Through the foregoing detailed description of the method for generating a sports training scheme based on limb feature recognition, those skilled in the art can clearly understand the system for generating a sports training scheme based on limb feature recognition in the embodiment. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and relevant parts are described in the method part.

[0073] The above description of disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for generating sports training programs based on limb feature recognition, characterized in that, include: Collect skeletal topology and body proportion profiles of the target users; Retrieve a list of joint numbers and a list of standard temporal coordinates for the training motion type that satisfy the skeletal topology profile and the body proportion profile; When the target user is training, the list of joint training time coordinates is extracted and compared with the list of standard joint time coordinates to obtain the list of joint time deviation coordinates. Iterate through the list of joint timing deviation coordinates and count the percentage of times when the coordinate deviation distance is greater than or equal to the deviation distance threshold. Based on the time percentage list, extract the set of joint numbers whose time percentage is greater than or equal to the time percentage threshold, input them into the action matching library to execute abnormal action configuration, obtain the abnormal action set, and configure the sports training plan. This includes collecting skeletal topology and body proportion profiles of the target users, including: Obtain the skeletal detection map and body shape detection map of the target user in the preset posture; The user receives the type of exercise to be trained, processes it through the active joint calibration library, and obtains the list of joint numbers. The active joint calibration library is pre-built by the user and stores a one-to-one correspondence between the exercise type and the list of joint numbers. Based on the joint number list, the distribution location of the bone detection map is identified to obtain joint topological features, which are then set as the bone topological portrait. Based on the joint number list, the limb proportion recognition is performed on the skeleton detection map to obtain limb proportion features, which are set as the body proportion portrait. The process includes retrieving a list of joint numbers and a list of standard temporal coordinates for the training motion type that satisfy the skeletal topology image and the body proportion image, including: Retrieve a group of qualified users with a first-level score who meet the skeletal topology profile and the body proportion profile, wherein any qualified user sample in the group of qualified users with a first-level score has a first-level skeletal topology profile and a first-level body proportion profile. When the number of qualified users in the first-level rating is less than or equal to the statistical threshold, the qualified users in the first-level rating are traversed, and the qualified users in the second-level rating who meet the first-level skeletal topology profile and the first-level body proportion profile are retrieved. Traverse the joint number list and perform high-frequency fitting of joint time-series coordinates on the first-level qualified user sample group and the second-level qualified user sample group to obtain the standard time-series coordinate list of joints.

2. The method as described in claim 1, characterized in that, Traverse the joint number list, perform high-frequency fitting of joint time-series coordinates on the first-level qualified user sample group and the second-level qualified user sample group to obtain the standard time-series coordinate list of joints, including: Extract the first joint number from the joint number list; Extract the first time-series coordinate set of the first joint number of the sample group of qualified users in the second-level rating; Extract the coordinate set from the first time step up to the coordinate set at the Nth time step from the time series coordinate set; Perform central tendency analysis by traversing the first time coordinate set up to the Nth time coordinate set to obtain the central values ​​of the first time coordinate set up to the Nth time coordinate set; Extract the second time-series coordinate set of the first joint number of the sample group of qualified users in the first-level rating; Traverse the coordinate set values ​​at the first time point up to the coordinate set values ​​at the Nth time point, perform simultaneous time-series trend analysis in conjunction with the second time-series coordinate set, obtain the standard time-series coordinates of the first joint number, and add them to the standard time-series coordinate list of the joint.

3. The method as described in claim 1, characterized in that, Based on the time percentage list, extract the set of joint numbers whose time percentage is greater than or equal to the time percentage threshold, input them into the action matching library to execute abnormal action configuration, obtain the abnormal action set, and configure a sports training plan, including: Step 1: Enumerate and combine k items of the joint number set to obtain several joint number combinations, where the initial value of k is equal to 1, the total number of joint numbers is ≥ k ≥ 1, and k is an integer; Step 2: Based on the type of movement to be trained, obtain the first combination of joint numbers up to the Mth combination of joint numbers from the movement matching library; Step 3: Traverse the several joint number combinations, match the first action joint number combination up to the Mth action joint number combination to obtain a set of k abnormal actions, and add them to the abnormal action set; If k < the total number of joint numbers, increment k by one and return to step one; otherwise, output the set of abnormal actions.

4. The method as described in claim 1, characterized in that, The action matching library update steps include: Collect several preset movement training logs of preset movement types, wherein the several preset movement training logs have several joint number record combinations; Based on the combination of the aforementioned joint number records, the trigger frequency percentage of the aforementioned joint numbers is statistically analyzed; Extract the joint numbers whose trigger frequency percentage is greater than or equal to the trigger frequency percentage threshold from the trigger frequency percentage of the plurality of joint numbers, add them to the preset action joint number combination and bind them to the preset motion type, and store them in the action matching library.

5. The method as described in claim 1, characterized in that, Based on the time percentage list, extract the set of joint numbers whose time percentage is greater than or equal to the time percentage threshold, input them into the action matching library to execute abnormal action configuration, obtain the abnormal action set, and configure a sports training plan, including: Based on the set of abnormal actions, extract the first abnormal action; Extract the number of times the first abnormal action has been trained; Based on historical training logs, the average number of training sessions for the first abnormal action was calculated. When the number of training iterations is less than the average number of training iterations, the target user is identified as a regular user. When the number of training iterations is greater than or equal to the average number of training iterations, the target user is identified for enhanced training. Configure a first abnormal action sports training scheme based on the regular user identifier or the enhanced training identifier, and add it to the sports training scheme.

6. A sports training program generation system based on limb feature recognition, characterized in that, A method for generating a sports training program based on limb feature recognition as described in any one of claims 1-5, the system comprising: The image acquisition module is used to acquire the skeletal topology image and body proportion image of the target user; The data retrieval module is used to retrieve a list of joint numbers and a list of standard temporal coordinates of joints that satisfy the skeletal topology profile and the body proportion profile for the type of exercise to be trained. The deviation comparison module is used to extract the joint training time coordinate list when the target user is training, and compare it with the joint standard time coordinate list to obtain the joint time deviation coordinate list. The statistical analysis module is used to traverse the list of joint time-series deviation coordinates and count the percentage of times when the coordinate deviation distance is greater than or equal to the deviation distance threshold. The scheme configuration module is used to extract a set of joint numbers whose time percentage is greater than or equal to the time percentage threshold based on the time percentage list, input the set of abnormal actions into the action matching library to execute abnormal action configuration, obtain the abnormal action set, and configure the sports training scheme.

Images